Advancing Spectrum Management with Machine Learning: A System Identification Case Study

Accurate system identification is essential for effective spectrum management, especially as the complexity and volume of spectrum usage increase. This paper presents a data-driven framework designed to facilitate and advance spectrum management activities such as database verification, frequency assignment, and compliance verification. The proposed approach utilizes diverse data sources, including crowdsourced spectrum data and direct measurements, and applies machine learning techniques to deliver scalable and reliable system identification solutions. Validation is performed through a case study in the Land Mobile Radio (LMR) bands using a prototype spectrum awareness system, which comprises a network of spectrum sensors and a cloud-based analysis platform. Empirical evaluation, encompassing both synthetic and real-world test cases, demonstrates the model's effectiveness in detecting discrepancies between spectrum management records and actual deployed systems.